The 2PC agent method and its correctness

A method for integrated concurrency control and recovery, applicable to heterogeneous multidatabase systems is proposed. The responsibility for two-phase local commitment and recovery of the prepared state at Participants is taken over by an entity called 2PC Agent. The main importance of the method is in preserving global serialisability in the presence of certain classes of Participant-related failures. The corresponding recovery methods, including a novel failure-time optimistic concurrency control method, are presented. The correctness of the 2PC Agent method is extensively studied. Rigorousness of the local histories is assumed throughout the proofs. Conflict serialisability is shown to hold for arbitrary histories consisting of local and global transactions, unless a unilateral abort happens to a global subtransaction that was in the prepared state. View serialisability is shown to hold for arbitrary histories where failures might happen at any time if the local transactions do not update the data accessed by the global transactions being in the prepared state. It is shown that the execution of any transaction is finite even if arbitrary Coordinator or site failures occur. Assuming additionally that any unilaterally aborted local subtransaction succeeds after a fixed number of resubmissions and that the Coordinator does not crash, it is shown that all transactions committed by the Coordinator eventually terminate. For globally aborted transactions it is shown that they terminate in a finite time if the local systems are not fully reliable.